feat: support kilo provider for judging

[shssoichiro]

Given that Kilo is supported as a benchmark provider in OpenClaw, and this software is from the Kilo team, it made sense to support Kilo as a judging provider. Requires KILO_API_KEY to be set in similar fashion to the other judging providers.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

This is a well-implemented addition. The _judge_via_kilo function follows the exact same pattern as _judge_via_openrouter — consistent with the existing codebase. The routing dispatch correctly inserts the kilo/ check before the other providers, and the new test file provides solid coverage (missing key, correct request shape, and dispatch isolation). Nothing to flag.

Files Reviewed (4 files)

• README.md
• scripts/benchmark.py
• scripts/lib_agent.py
• tests/test_lib_agent_judge.py

---

Fix these issues in Kilo Cloud

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

This is a clean, well-structured addition. The _judge_via_kilo implementation follows the exact same pattern as the existing _judge_via_openrouter and _judge_via_openai providers — consistent error handling, KILO_API_KEY guard, model prefix stripping, and delegation to the shared _judge_via_openai_compat helper. The new test file covers the key cases (missing key, correct URL/payload, dispatch isolation) thoroughly.

Files Reviewed (4 files)

• README.md
• scripts/benchmark.py
• scripts/lib_agent.py
• tests/test_lib_agent_judge.py

---

_{Reviewed by claude-sonnet-4.6 · 119,399 tokens}